Socratic-Geo: Synthetic Data Generation and Geometric Reasoning via Multi-Agent Interaction
Zhengbo Jiao, Shaobo Wang, Zifan Zhang, Wei Wang, Bing Zhao, Hu Wei, Linfeng Zhang · 2026
Read Socratic-Geo: Synthetic Data Generation and Geometric Reasoning via Multi-Agent Interaction by Zhengbo Jiao, Shaobo Wang, Zifan Zhang, Wei Wang, Bing Zhao, Hu Wei, Linfeng Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03414.txt.
Description
Geometry problems are scarce training material for AI, because good image-text pairs are expensive to annotate by hand. Zhengbo Jiao and 6 co-authors answer that shortage with Socratic-Geo, a 2026 framework where several agents build data and learn from it together. A Teacher agent writes Python scripts that draw the figures and pose the problems.
Feedback loops check each generated problem for solvability and visual validity before it enters training. A Solver agent learns by preference training, and its failures tell the Teacher what to generate next. A separate Generator learns image creation from the accumulated code triplets. Starting from just 108 seed problems, the system beats strong baselines on six benchmarks.
Catalog record
| CREATOR | Zhengbo Jiao, Shaobo Wang, Zifan Zhang, Wei Wang, Bing Zhao, Hu Wei, Linfeng Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03414